Classification results of the test set.

Results for the test set, including the percentage of correctly classified metabolites (Sensitivity), the percentage of correctly classified non-metabolites (Specificity) and the Area Under the Curve (AUC). It can be observed that the best combination of descriptor and classifier is MDL Public Keys and Random Forest and that the second best is ECFP_4 fingerprints and Random Forest. Interestingly, physicochemical descriptors (PP_desc) perform well both with Random Forest and Support Vector Machines classifiers. (A molecule is considered metabolite if its metabolite-likeness >50%).